Episodic Memory
Generation Effect Task
Also called: self-generation effect, generate versus read, read versus generate, word fragment generation
Type the second word of English word pairs, some shown whole and some with letters to work out, then recognize them among new words. Measures how much generating a word helps memory.
8-10 minUntimed
Free to start: 10 free participants per account, then $0.99 per participant for academic research, or any yearly plan. Try it here first, with no account.
- Papers, 2021 to 2025
- 96
- Evidence grade
- C
- Registered clinical studies
- 4
What it measures
Measures the memory benefit of generating a word. In one list of English word pairs, some second words are shown whole and others with letters missing, to be worked out from the first word; every second word is typed. A recognition test then shows them among new words.
Constructs
- Generation effect
- Encoding
- Recognition memory
- Episodic memory
Research uses
- Memory and learning research
- Education research on study strategies
- Aging and memory research
How the paradigm works
The Generation Effect Task measures how much generating a word helps people remember it. Slamecka and Graf (1978) named the effect: words that people produce themselves from a cue are remembered better than the same words read. In their study each item gave a first word and a rule, and the second word was either shown whole or left for the participant to produce from its first letter; both kinds were said aloud, and the generated words were recognized better under every rule they tried.
More about the paradigm
In this version, one list mixes pairs of English words whose second word is shown whole with pairs whose second word has letters missing, one blank for each, to be worked out from the first word: hot, c_ld. The two words of a pair are opposites, words of nearly the same meaning, or things that belong together. The participant types the second word in full for every pair and sends it with Enter or Next, so the two kinds differ only in working the word out. Each pair stays until its word is sent. After 30 seconds of true-or-false arithmetic, a recognition test shows every second word among as many new words, one at a time, answered Yes or No with no time limit. The standard list has 16 pairs of each kind and the test 64 words; a setting allows 10 to 20 of each kind. The pairs were written for this task in English, and each session gives every pair its role at random.
The main score is the generation effect: the hit rate for generated words minus the hit rate for read words. As Slamecka and Graf did, a word typed wrong at study is left out of its kind's hit rate; the effect over every studied word, typed right or not, is reported too. The hit rates, the false alarm rate for new words, corrected recognition and d′ for each kind, the share of words typed right, and the median time each kind took are also reported. In McCurdy and colleagues' (2020) meta-analysis, generated items were remembered better than read ones by about 0.14 in proportion remembered, both in mixed lists and on recognition tests, the two features of this version. One person's effect over 16 words of each kind is noisy, so it is best used to compare groups. The words are English in every session, and sessions run in another language are flagged. Scores support research and do not diagnose.
Scores
- Generation Effect%
- Hit rate for generated words minus hit rate for read words, among words typed right.
- Positive means generated words were recognized more often
- Hit Rate, Generated Words%
- Share of generated words answered Yes, among words typed right.
- Higher is better
- Hit Rate, Read Words%
- Share of read words answered Yes, among words typed right.
- Higher is better
- False Alarm Rate%
- Share of new words answered Yes.
- Lower is better
- Corrected Recognition, Generated Words%
- Generated hit rate minus the false alarm rate.
- Higher is better
- Corrected Recognition, Read Words%
- Read hit rate minus the false alarm rate.
- Higher is better
- d′, Generated Wordsd'
- z(generated hit rate) minus z(false alarm rate), with the log-linear correction.
- Higher is better
- d′, Read Wordsd'
- z(read hit rate) minus z(false alarm rate), with the log-linear correction.
- Higher is better
- Recognition Accuracy%
- Test answers right: Yes for studied words, No for new ones.
- Higher is better
- Words Worked Out%
- Share of words with missing letters typed right, allowing one letter off in words of five letters or more.
- Shows the generating was done
- Words Copied%
- Share of words shown whole typed right, allowing one letter off in words of five letters or more.
- Shows the typing was done
- Generation Effect, All Words%
- The generation effect over every studied word, typed right or not.
- Positive means generated words were recognized more often
- Time per Generated Wordms
- Median time from a pair with missing letters appearing to its word being sent.
- Context for the effect
- Time per Read Wordms
- Median time from a whole pair appearing to its word being sent.
- Context for the effect
Use and evidence
96 papers from 2021 to 2025 name this task in their title or abstract, down 9% compared with 2016 to 2020. That ranks it 222 of 495 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 33 |
| 2024 | 11 |
| 2023 | 16 |
| 2022 | 14 |
| 2021 | 22 |
| 2020 | 20 |
| 2019 | 25 |
| 2018 | 23 |
| 2017 | 13 |
| 2016 | 25 |
| 2015 | 11 |
| 2014 | 13 |
| 2013 | 12 |
| 2012 | 22 |
| 2011 | 16 |
| 2010 | 20 |
| 2009 | 16 |
| 2008 | 14 |
| 2007 | 16 |
| 2006 | 12 |
| 2005 | 13 |
| 2004 | 14 |
| 2003 | 8 |
| 2002 | 13 |
| 2001 | 10 |
| 2000 | 11 |
Use grade D
30 to 99 papers from 2021 to 2025.
Evidence grade C
At least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 5
- Registered clinical studies using it as an outcome
- 4
Note on the counts: a few counted papers use the words for a difference between generations; its 4 trial listings are off-topic.
How the grades work
Use: A is 1,000 or more papers from 2021 to 2025, B 300 to 999, C 100 to 299, D 30 to 99 and E fewer than 30. Evidence: A means at least 25 papers report its test-retest reliability and at least 25 registered clinical studies use it as an outcome; B, at least 8 test-retest papers, and either 5 registered clinical studies or 150 reliability or validity papers; C, at least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies; D, little published measurement evidence found. The counts measure use. They include every paper that names the task in its title or abstract. Grades describe the published research on this task, not ConductCognition's version of it.
Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Study
- One list of 32 English word pairs (standard; 20 to 40): 16 with the second word to work out from missing letters, one blank each (standard; 10 to 20 of each kind), and 16 with it shown whole, mixed with no more than 4 of a kind in a row. The second word is typed in full for every pair and sent with Enter or Next; each pair stays until then, with a 500 ms gap after.
- Delay
- 30 seconds (standard; 0 to 120) of true-or-false arithmetic.
- Test
- Every second word among as many new words, one at a time, answered Yes or No with no time limit and no feedback; no more than 4 studied or new words in a row.
- Words
- Opposites, words of nearly the same meaning and words that belong together, written for this task in English. Each session gives every pair its role at random, and the words are English in every session.
- Scores
- The generation effect: the hit rate for generated words minus the hit rate for read words, among words typed right. Hit and false alarm rates, corrected recognition and d′ for each kind, the share of words typed right, and the effect over every word.
- Practice
- 4 pairs, 2 of each kind, before the list.
Slamecka and Graf (1978) found that words people produce from a cue are remembered better than the same words read, under every rule they tried. Here both kinds of word are typed, so the two differ only in working the word out. The effect is larger when generated and read words are mixed in one list, as here, and one person's score is noisy, so compare groups. The word pairs were written for this task; no published list is used. The scores support research and do not diagnose.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: generation_effect_test from the scored run, one Yes or No answer per test word, half of them studied and half new, with the study list and the delay arithmetic from the same run. Accuracy significantly below chance (one-sided exact binomial, p < .05, chance 50%) invalidates; accuracy not above chance over 10 or more answers is flagged. The same answer on 90% of the words invalidates when accuracy is not above chance. Answers under 200 ms on at least half of 4 or more answered words invalidate. Fewer than 90% of the planned test words answered invalidates. Fewer than half the words with missing letters typed right is flagged and withholds the percentile. No arithmetic answered in the delay, or delay arithmetic at or below chance over at least 10 answers, is flagged. A session run in another language than English is flagged. A hidden tab withholds the percentile.
Why these rules
Half the test words were studied and half are new, so guessing gives 50%: Accuracy significantly below chance (one-sided exact test, p < .05) cannot come from guessing; it means the response mapping was reversed or the task was not followed (Slick, Sherman and Iverson 1999 treat below-chance forced-choice performance as definite response bias). Chance-level accuracy can be a real result in impaired participants, so it is only flagged. One option on at least 90% of answered trials, when the correct options are balanced and accuracy is not above chance, means the answers did not depend on the items (the task counterpart of long-string responding in surveys; Curran 2016). An answer under 200 ms is faster than a simple visual reaction, so it cannot reflect the item. Following response-time effort screening for untimed tests (Wise and Kong 2005), the session is invalid only when at least half the answers are that fast. The instructions ask for Yes on every word typed in the list, whole or worked out, so accuracy is judged over both kinds; the effect itself, which may be zero or negative, is never judged. When most words with missing letters were typed wrong, the generated words' hit rate rests on few words. Skipping the delay arithmetic means that part of the procedure may not have been followed. The word lists are English. A session run in another language shows English words to a participant who may not read English fluently, so it is flagged for review. A hidden tab lets the study, delay or response timers run without the participant, so a percentile is withheld after one.
No percentile. Scores and quality flags only.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Pairs of each kind | 16 | 10 to 20 |
| Delay before the test (seconds) | 30 | 0 to 120 |
Yearly plans can change settings within these limits. Free and pay-as-you-go studies use the standard settings.
Need a setting this test does not have?
Devices and languages
- Untimed test: runs on desktops, laptops, tablets and phones. Participant devices
- Instructions and screens: English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese. Words and sentences: English.
- Available on: Every plan, including the free plan.
- No percentile. Scores and quality flags only. How norms work
- Recruiting online? See the setup guides for Prolific, Sona Systems, CloudResearch Connect and Qualtrics.
Cite and describe
Participants completed the Generation Effect Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Slamecka & Graf, 1978), with the standard settings (pairs of each kind: 16; delay before the test (seconds): 30). The main score was generation effect (hit rate for generated words minus hit rate for read words, among words typed right).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592-604. https://doi.org/10.1037/0278-7393.4.6.592The generation effect: words produced from a rule, a cue and a first letter, against the same words read.
- McCurdy, M. P., Viechtbauer, W., Sklenar, A. M., Frankenstein, A. N., & Leshikar, E. D. (2020). Theories of the generation effect and the impact of generation constraint: A meta-analytic review. Psychonomic Bulletin & Review, 27(6), 1139-1165. https://doi.org/10.3758/s13423-020-01762-3The effect's size by list design, memory test and generation task.
Source papers
- Slamecka NJ, Graf P (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6):592-604.Paradigm source
- McCurdy MP, Viechtbauer W, Sklenar AM, Frankenstein AN, Leshikar ED (2020). Theories of the generation effect and the impact of generation constraint: A meta-analytic review. Psychonomic Bulletin & Review, 27(6):1139-1165.Paradigm source
Questions
How long does the Generation Effect Task take?
About 8 to 10 minutes, as listed in the test library.
What device does a participant need?
Untimed test: runs on desktops, laptops, tablets and phones. The browser must be Chrome, Edge, Firefox or Safari.
What scores does it produce?
Generation Effect, Hit Rate, Generated Words, Hit Rate, Read Words, False Alarm Rate, Corrected Recognition, Generated Words, Corrected Recognition, Read Words, d′, Generated Words, d′, Read Words, Recognition Accuracy, Words Worked Out, Words Copied, Generation Effect, All Words, Time per Generated Word and Time per Read Word. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Generation Effect Task used in research?
96 papers from 2021 to 2025 name it in their title or abstract, down 9% compared with 2016 to 2020. That ranks it 222 of 495 test families in the ConductCognition test census (use grade D). Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Generation Effect Task have?
In the published research, 0 papers report its test-retest reliability and 4 registered clinical studies use it as an outcome (evidence grade C in the ConductCognition test census).
Does it report a percentile?
No. Scores and quality flags only.
Which settings can a researcher change?
Pairs of each kind (10 to 20) and delay before the test (seconds) (0 to 120). Settings can be changed on yearly plans; free and pay-as-you-go studies use the standard settings.
Which languages can participants use?
The instructions and screens are in English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese. The words and sentences stay in English, so sessions in other languages are flagged for review. English is on every plan, and the other languages come with yearly plans. Have a native speaker read the machine-translated text before launch.
Is this a clinical test?
No. ConductCognition is for research use only. It runs, scores and exports the test; clinical interpretation is outside the platform.
Every test: the full test list.
Run this test in your next study
Start free with 10 free participants on any test, then use participant credits or a yearly plan.